static void brw_set_src0( struct brw_instruction *insn,
struct brw_reg reg )
{
- assert(reg.file != BRW_MESSAGE_REGISTER_FILE);
-
if (reg.type != BRW_ARCHITECTURE_REGISTER_FILE)
assert(reg.nr < 128);
struct intel_context *intel = &brw->intel;
brw_set_src1(insn, brw_imm_d(0));
- if (intel->is_ironlake) {
- insn->bits3.math_igdng.function = function;
- insn->bits3.math_igdng.int_type = integer_type;
- insn->bits3.math_igdng.precision = low_precision;
- insn->bits3.math_igdng.saturate = saturate;
- insn->bits3.math_igdng.data_type = dataType;
- insn->bits3.math_igdng.snapshot = 0;
- insn->bits3.math_igdng.header_present = 0;
- insn->bits3.math_igdng.response_length = response_length;
- insn->bits3.math_igdng.msg_length = msg_length;
- insn->bits3.math_igdng.end_of_thread = 0;
- insn->bits2.send_igdng.sfid = BRW_MESSAGE_TARGET_MATH;
- insn->bits2.send_igdng.end_of_thread = 0;
+ if (intel->gen == 5) {
+ insn->bits3.math_gen5.function = function;
+ insn->bits3.math_gen5.int_type = integer_type;
+ insn->bits3.math_gen5.precision = low_precision;
+ insn->bits3.math_gen5.saturate = saturate;
+ insn->bits3.math_gen5.data_type = dataType;
+ insn->bits3.math_gen5.snapshot = 0;
+ insn->bits3.math_gen5.header_present = 0;
+ insn->bits3.math_gen5.response_length = response_length;
+ insn->bits3.math_gen5.msg_length = msg_length;
+ insn->bits3.math_gen5.end_of_thread = 0;
+ insn->bits2.send_gen5.sfid = BRW_MESSAGE_TARGET_MATH;
+ insn->bits2.send_gen5.end_of_thread = 0;
} else {
insn->bits3.math.function = function;
insn->bits3.math.int_type = integer_type;
{
brw_set_src1(insn, brw_imm_d(0));
- insn->bits3.urb_igdng.opcode = 1;
- insn->bits3.urb_igdng.offset = offset;
- insn->bits3.urb_igdng.swizzle_control = swizzle_control;
- insn->bits3.urb_igdng.allocate = allocate;
- insn->bits3.urb_igdng.used = used;
- insn->bits3.urb_igdng.complete = complete;
- insn->bits3.urb_igdng.header_present = 1;
- insn->bits3.urb_igdng.response_length = response_length;
- insn->bits3.urb_igdng.msg_length = msg_length;
- insn->bits3.urb_igdng.end_of_thread = end_of_thread;
- insn->bits2.send_igdng.sfid = BRW_MESSAGE_TARGET_URB;
- insn->bits2.send_igdng.end_of_thread = end_of_thread;
+ insn->bits3.urb_gen5.opcode = 1;
+ insn->bits3.urb_gen5.offset = offset;
+ insn->bits3.urb_gen5.swizzle_control = swizzle_control;
+ insn->bits3.urb_gen5.allocate = allocate;
+ insn->bits3.urb_gen5.used = used;
+ insn->bits3.urb_gen5.complete = complete;
+ insn->bits3.urb_gen5.header_present = 1;
+ insn->bits3.urb_gen5.response_length = response_length;
+ insn->bits3.urb_gen5.msg_length = msg_length;
+ insn->bits3.urb_gen5.end_of_thread = end_of_thread;
+ insn->bits2.send_gen5.sfid = BRW_MESSAGE_TARGET_URB;
+ insn->bits2.send_gen5.end_of_thread = end_of_thread;
}
static void brw_set_urb_message( struct brw_context *brw,
struct intel_context *intel = &brw->intel;
brw_set_src1(insn, brw_imm_d(0));
- if (intel->is_ironlake) {
- insn->bits3.urb_igdng.opcode = 0; /* ? */
- insn->bits3.urb_igdng.offset = offset;
- insn->bits3.urb_igdng.swizzle_control = swizzle_control;
- insn->bits3.urb_igdng.allocate = allocate;
- insn->bits3.urb_igdng.used = used; /* ? */
- insn->bits3.urb_igdng.complete = complete;
- insn->bits3.urb_igdng.header_present = 1;
- insn->bits3.urb_igdng.response_length = response_length;
- insn->bits3.urb_igdng.msg_length = msg_length;
- insn->bits3.urb_igdng.end_of_thread = end_of_thread;
- insn->bits2.send_igdng.sfid = BRW_MESSAGE_TARGET_URB;
- insn->bits2.send_igdng.end_of_thread = end_of_thread;
+ if (intel->gen >= 5) {
+ insn->bits3.urb_gen5.opcode = 0; /* ? */
+ insn->bits3.urb_gen5.offset = offset;
+ insn->bits3.urb_gen5.swizzle_control = swizzle_control;
+ insn->bits3.urb_gen5.allocate = allocate;
+ insn->bits3.urb_gen5.used = used; /* ? */
+ insn->bits3.urb_gen5.complete = complete;
+ insn->bits3.urb_gen5.header_present = 1;
+ insn->bits3.urb_gen5.response_length = response_length;
+ insn->bits3.urb_gen5.msg_length = msg_length;
+ insn->bits3.urb_gen5.end_of_thread = end_of_thread;
+ if (intel->gen >= 6) {
+ /* For SNB, the SFID bits moved to the condmod bits, and
+ * EOT stayed in bits3 above. Does the EOT bit setting
+ * below on Ironlake even do anything?
+ */
+ insn->header.destreg__conditionalmod = BRW_MESSAGE_TARGET_URB;
+ } else {
+ insn->bits2.send_gen5.sfid = BRW_MESSAGE_TARGET_URB;
+ insn->bits2.send_gen5.end_of_thread = end_of_thread;
+ }
} else {
insn->bits3.urb.opcode = 0; /* ? */
insn->bits3.urb.offset = offset;
struct intel_context *intel = &brw->intel;
brw_set_src1(insn, brw_imm_d(0));
- if (intel->is_ironlake) {
- insn->bits3.dp_write_igdng.binding_table_index = binding_table_index;
- insn->bits3.dp_write_igdng.msg_control = msg_control;
- insn->bits3.dp_write_igdng.pixel_scoreboard_clear = pixel_scoreboard_clear;
- insn->bits3.dp_write_igdng.msg_type = msg_type;
- insn->bits3.dp_write_igdng.send_commit_msg = 0;
- insn->bits3.dp_write_igdng.header_present = 1;
- insn->bits3.dp_write_igdng.response_length = response_length;
- insn->bits3.dp_write_igdng.msg_length = msg_length;
- insn->bits3.dp_write_igdng.end_of_thread = end_of_thread;
- insn->bits2.send_igdng.sfid = BRW_MESSAGE_TARGET_DATAPORT_WRITE;
- insn->bits2.send_igdng.end_of_thread = end_of_thread;
+ if (intel->gen == 5) {
+ insn->bits3.dp_write_gen5.binding_table_index = binding_table_index;
+ insn->bits3.dp_write_gen5.msg_control = msg_control;
+ insn->bits3.dp_write_gen5.pixel_scoreboard_clear = pixel_scoreboard_clear;
+ insn->bits3.dp_write_gen5.msg_type = msg_type;
+ insn->bits3.dp_write_gen5.send_commit_msg = 0;
+ insn->bits3.dp_write_gen5.header_present = 1;
+ insn->bits3.dp_write_gen5.response_length = response_length;
+ insn->bits3.dp_write_gen5.msg_length = msg_length;
+ insn->bits3.dp_write_gen5.end_of_thread = end_of_thread;
+ insn->bits2.send_gen5.sfid = BRW_MESSAGE_TARGET_DATAPORT_WRITE;
+ insn->bits2.send_gen5.end_of_thread = end_of_thread;
} else {
insn->bits3.dp_write.binding_table_index = binding_table_index;
insn->bits3.dp_write.msg_control = msg_control;
struct intel_context *intel = &brw->intel;
brw_set_src1(insn, brw_imm_d(0));
- if (intel->is_ironlake) {
- insn->bits3.dp_read_igdng.binding_table_index = binding_table_index;
- insn->bits3.dp_read_igdng.msg_control = msg_control;
- insn->bits3.dp_read_igdng.msg_type = msg_type;
- insn->bits3.dp_read_igdng.target_cache = target_cache;
- insn->bits3.dp_read_igdng.header_present = 1;
- insn->bits3.dp_read_igdng.response_length = response_length;
- insn->bits3.dp_read_igdng.msg_length = msg_length;
- insn->bits3.dp_read_igdng.pad1 = 0;
- insn->bits3.dp_read_igdng.end_of_thread = end_of_thread;
- insn->bits2.send_igdng.sfid = BRW_MESSAGE_TARGET_DATAPORT_READ;
- insn->bits2.send_igdng.end_of_thread = end_of_thread;
+ if (intel->gen == 5) {
+ insn->bits3.dp_read_gen5.binding_table_index = binding_table_index;
+ insn->bits3.dp_read_gen5.msg_control = msg_control;
+ insn->bits3.dp_read_gen5.msg_type = msg_type;
+ insn->bits3.dp_read_gen5.target_cache = target_cache;
+ insn->bits3.dp_read_gen5.header_present = 1;
+ insn->bits3.dp_read_gen5.response_length = response_length;
+ insn->bits3.dp_read_gen5.msg_length = msg_length;
+ insn->bits3.dp_read_gen5.pad1 = 0;
+ insn->bits3.dp_read_gen5.end_of_thread = end_of_thread;
+ insn->bits2.send_gen5.sfid = BRW_MESSAGE_TARGET_DATAPORT_READ;
+ insn->bits2.send_gen5.end_of_thread = end_of_thread;
} else {
insn->bits3.dp_read.binding_table_index = binding_table_index; /*0:7*/
insn->bits3.dp_read.msg_control = msg_control; /*8:11*/
assert(eot == 0);
brw_set_src1(insn, brw_imm_d(0));
- if (intel->is_ironlake) {
- insn->bits3.sampler_igdng.binding_table_index = binding_table_index;
- insn->bits3.sampler_igdng.sampler = sampler;
- insn->bits3.sampler_igdng.msg_type = msg_type;
- insn->bits3.sampler_igdng.simd_mode = simd_mode;
- insn->bits3.sampler_igdng.header_present = header_present;
- insn->bits3.sampler_igdng.response_length = response_length;
- insn->bits3.sampler_igdng.msg_length = msg_length;
- insn->bits3.sampler_igdng.end_of_thread = eot;
- insn->bits2.send_igdng.sfid = BRW_MESSAGE_TARGET_SAMPLER;
- insn->bits2.send_igdng.end_of_thread = eot;
+ if (intel->gen == 5) {
+ insn->bits3.sampler_gen5.binding_table_index = binding_table_index;
+ insn->bits3.sampler_gen5.sampler = sampler;
+ insn->bits3.sampler_gen5.msg_type = msg_type;
+ insn->bits3.sampler_gen5.simd_mode = simd_mode;
+ insn->bits3.sampler_gen5.header_present = header_present;
+ insn->bits3.sampler_gen5.response_length = response_length;
+ insn->bits3.sampler_gen5.msg_length = msg_length;
+ insn->bits3.sampler_gen5.end_of_thread = eot;
+ insn->bits2.send_gen5.sfid = BRW_MESSAGE_TARGET_SAMPLER;
+ insn->bits2.send_gen5.end_of_thread = eot;
} else if (intel->is_g4x) {
insn->bits3.sampler_g4x.binding_table_index = binding_table_index;
insn->bits3.sampler_g4x.sampler = sampler;
ALU2(DP3)
ALU2(DP2)
ALU2(LINE)
-
+ALU2(PLN)
struct brw_instruction *insn;
GLuint br = 1;
- if (intel->is_ironlake)
+ if (intel->gen == 5)
br = 2;
if (p->single_program_flow) {
struct intel_context *intel = &p->brw->intel;
GLuint br = 1;
- if (intel->is_ironlake)
+ if (intel->gen == 5)
br = 2;
if (p->single_program_flow) {
struct brw_instruction *insn;
GLuint br = 1;
- if (intel->is_ironlake)
+ if (intel->gen == 5)
br = 2;
if (p->single_program_flow)
struct brw_instruction *landing = &p->store[p->nr_insn];
GLuint jmpi = 1;
- if (intel->is_ironlake)
+ if (intel->gen == 5)
jmpi = 2;
assert(jmp_insn->header.opcode == BRW_OPCODE_JMPI);
GLuint data_type,
GLuint precision )
{
- struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
- GLuint msg_length = (function == BRW_MATH_FUNCTION_POW) ? 2 : 1;
- GLuint response_length = (function == BRW_MATH_FUNCTION_SINCOS) ? 2 : 1;
+ struct intel_context *intel = &p->brw->intel;
- /* Example code doesn't set predicate_control for send
- * instructions.
- */
- insn->header.predicate_control = 0;
- insn->header.destreg__conditionalmod = msg_reg_nr;
+ if (intel->gen >= 6) {
+ struct brw_instruction *insn = next_insn(p, BRW_OPCODE_MATH);
- brw_set_dest(insn, dest);
- brw_set_src0(insn, src);
- brw_set_math_message(p->brw,
- insn,
- msg_length, response_length,
- function,
- BRW_MATH_INTEGER_UNSIGNED,
- precision,
- saturate,
- data_type);
+ /* Math is the same ISA format as other opcodes, except that CondModifier
+ * becomes FC[3:0] and ThreadCtrl becomes FC[5:4].
+ */
+ insn->header.destreg__conditionalmod = function;
+
+ brw_set_dest(insn, dest);
+ brw_set_src0(insn, src);
+ brw_set_src1(insn, brw_null_reg());
+ } else {
+ struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
+ GLuint msg_length = (function == BRW_MATH_FUNCTION_POW) ? 2 : 1;
+ GLuint response_length = (function == BRW_MATH_FUNCTION_SINCOS) ? 2 : 1;
+ /* Example code doesn't set predicate_control for send
+ * instructions.
+ */
+ insn->header.predicate_control = 0;
+ insn->header.destreg__conditionalmod = msg_reg_nr;
+
+ brw_set_dest(insn, dest);
+ brw_set_src0(insn, src);
+ brw_set_math_message(p->brw,
+ insn,
+ msg_length, response_length,
+ function,
+ BRW_MATH_INTEGER_UNSIGNED,
+ precision,
+ saturate,
+ data_type);
+ }
}
/**
GLuint simd_mode)
{
GLboolean need_stall = 0;
-
+
if (writemask == 0) {
/*printf("%s: zero writemask??\n", __FUNCTION__); */
return;
/* printf("need stall %x %x\n", newmask , writemask); */
}
else {
+ GLboolean dispatch_16 = GL_FALSE;
+
struct brw_reg m1 = brw_message_reg(msg_reg_nr);
-
+
+ guess_execution_size(p->current, dest);
+ if (p->current->header.execution_size == BRW_EXECUTE_16)
+ dispatch_16 = GL_TRUE;
+
newmask = ~newmask & WRITEMASK_XYZW;
brw_push_insn_state(p);
src0 = retype(brw_null_reg(), BRW_REGISTER_TYPE_UW);
dest = offset(dest, dst_offset);
- response_length = len * 2;
+
+ /* For 16-wide dispatch, masked channels are skipped in the
+ * response. For 8-wide, masked channels still take up slots,
+ * and are just not written to.
+ */
+ if (dispatch_16)
+ response_length = len * 2;
}
}
GLuint offset,
GLuint swizzle)
{
- struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
+ struct intel_context *intel = &p->brw->intel;
+ struct brw_instruction *insn;
+
+ /* Sandybridge doesn't have the implied move for SENDs,
+ * and the first message register index comes from src0.
+ */
+ if (intel->gen >= 6) {
+ brw_push_insn_state(p);
+ brw_set_mask_control( p, BRW_MASK_DISABLE );
+ brw_MOV(p, brw_message_reg(msg_reg_nr), src0);
+ brw_pop_insn_state(p);
+ src0 = brw_message_reg(msg_reg_nr);
+ }
+
+ insn = next_insn(p, BRW_OPCODE_SEND);
assert(msg_length < BRW_MAX_MRF);
brw_set_src0(insn, src0);
brw_set_src1(insn, brw_imm_d(0));
- insn->header.destreg__conditionalmod = msg_reg_nr;
+ if (intel->gen < 6)
+ insn->header.destreg__conditionalmod = msg_reg_nr;
brw_set_urb_message(p->brw,
insn,